[PATCH 6.12.y 0/2] spi: cadence-quadspi: fix probe error path and remove

[PATCH 6.12.y 0/2] spi: cadence-quadspi: fix probe error path and remove

[Diogo Ivo (Schneider Electric)]

This series fixes two resource leaks in the cadence-quadspi driver's
probe error path and remove callback.

Patch 1 fixes an imbalanced pm_runtime_enable()/pm_runtime_disable()
pair: when probe defers after cqspi_request_mmap_dma(),
pm_runtime_disable() is called without a prior pm_runtime_enable().
It also adds the missing pm_runtime_dont_use_autosuspend() in both
the probe error path and the remove callback and a missing
pm_runtime_put_noidle() in the probe error path. This consolidates
fixes from upstream commits 5ff4d5d1af0c and 5e8bb0cc72f1.

Patch 2 fixes a DMA channel leak: when spi_register_controller() fails,
the channel acquired via cqspi_request_mmap_dma() is never released.
A dedicated error label is added to release it before the runtime PM
teardown. This fix is one part of upstream commit f18c8cfa4f1a.

Signed-off-by: Diogo Ivo (Schneider Electric) <diogo.ivo@bootlin.com>
---
Diogo Ivo (Schneider Electric) (2):
      spi: cadence-quadspi: fix runtime pm cleanup on probe failure/unbind
      spi: cadence-quadspi: release DMA channel on probe failure

 drivers/spi/spi-cadence-quadspi.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
---
base-commit: 7351870f56cba79097a2ab57e609469719ea4f73
change-id: 20260730-cqspi-pm_runtime-a8a1c6122944

Best regards,
--  
Diogo Ivo <diogo.ivo@bootlin.com>

[PATCH 6.12.y 1/2] spi: cadence-quadspi: fix runtime pm cleanup on probe failure/unbind

[Diogo Ivo (Schneider Electric)]

Currently the probe cleanup path is imbalanced regarding
pm_runtime_enable() and pm_runtime_disable(). For example, if the probe
defers after calling cqspi_request_mmap_dma() pm_runtime_disable() will
be called without a prior call to pm_runtime_enable(), causing an
imbalance. Furthermore, the cleanup path does not undo all the runtime
calls made while probing.

Fix this by enforcing the correct correspondence between
pm_runtime_enable() and pm_runtime_resume() and add the missing PM
cleanup calls.

As the history of this driver in mainline is convoluted with several
rounds of fixes it includes the fixes from two commits, namely
commit 5ff4d5d1af0c ("spi: cadence-quadspi: fix runtime pm disable
imbalance on probe failure") and commit 5e8bb0cc72f1 ("spi:
cadence-quadspi: fix runtime pm and clock imbalance on unbind").

Fixes: 74b0b4cf13fc ("spi: spi-cadence-quadspi: Fix pm runtime unbalance")
Cc: stable@vger.kernel.org
Signed-off-by: Diogo Ivo (Schneider Electric) <diogo.ivo@bootlin.com>
---
 drivers/spi/spi-cadence-quadspi.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/spi/spi-cadence-quadspi.c b/drivers/spi/spi-cadence-quadspi.c
index da8401261bbc..d749dee54015 100644
--- a/drivers/spi/spi-cadence-quadspi.c
+++ b/drivers/spi/spi-cadence-quadspi.c
@@ -1862,7 +1862,6 @@ static int cqspi_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
-
 	ret = clk_prepare_enable(cqspi->clk);
 	if (ret) {
 		dev_err(dev, "Cannot enable QSPI clock.\n");
@@ -1983,16 +1982,19 @@ static int cqspi_probe(struct platform_device *pdev)
 	ret = spi_register_controller(host);
 	if (ret) {
 		dev_err(&pdev->dev, "failed to register SPI ctlr %d\n", ret);
-		goto probe_setup_failed;
+		goto disable_rpm;
 	}
 
 	pm_runtime_mark_last_busy(dev);
 	pm_runtime_put_autosuspend(dev);
 
 	return 0;
+disable_rpm:
+	pm_runtime_put_noidle(dev);
+	pm_runtime_dont_use_autosuspend(dev);
+	pm_runtime_disable(dev);
 probe_setup_failed:
 	cqspi_controller_enable(cqspi, 0);
-	pm_runtime_disable(dev);
 probe_reset_failed:
 	if (cqspi->is_jh7110)
 		cqspi_jh7110_disable_clk(pdev, cqspi);
@@ -2026,6 +2028,7 @@ static void cqspi_remove(struct platform_device *pdev)
 		cqspi_jh7110_disable_clk(pdev, cqspi);
 
 	pm_runtime_put_sync(&pdev->dev);
+	pm_runtime_dont_use_autosuspend(&pdev->dev);
 	pm_runtime_disable(&pdev->dev);
 }
 

-- 
2.55.0

Re: [PATCH 6.12.y 1/2] spi: cadence-quadspi: fix runtime pm cleanup on probe failure/unbind

[Sasha Levin]

On Fri, Aug 07, 2026 at 01:49:20PM +0200, Diogo Ivo (Schneider Electric) wrote:
> +disable_rpm:
> +	pm_runtime_put_noidle(dev);
> +	pm_runtime_dont_use_autosuspend(dev);
> +	pm_runtime_disable(dev);
>  probe_setup_failed:
>  	cqspi_controller_enable(cqspi, 0);
> -	pm_runtime_disable(dev);

This teardown order is inverted relative to upstream, and it reinstates
a bug upstream deliberately closed.

pm_runtime_dont_use_autosuspend() runs update_autosuspend(), and with
the usage count already dropped to 0 by the preceding put_noidle() and
the device still RPM_ACTIVE, that can queue a runtime suspend, i.e.
cqspi_runtime_suspend() -> clk_disable_unprepare(). The following
pm_runtime_disable() does barrier it, but control then falls through to
probe_setup_failed, which does cqspi_controller_enable(cqspi, 0) - a
register write against a potentially gated clock. That is exactly the
unclocked access 233db2cb14db ("spi: cadence-quadspi: fix unclocked
access on unbind") went out of its way to prevent.

Please mirror the upstream ordering from 5e8bb0cc72f1 ("spi:
cadence-quadspi: fix runtime pm and clock imbalance on unbind"):

	pm_runtime_disable(dev);
	pm_runtime_set_suspended(dev);
	pm_runtime_put_noidle(dev);
	pm_runtime_dont_use_autosuspend(dev);

pm_runtime_set_suspended() is missing here as well.

>  	pm_runtime_put_sync(&pdev->dev);
> +	pm_runtime_dont_use_autosuspend(&pdev->dev);
>  	pm_runtime_disable(&pdev->dev);

Same ordering problem in cqspi_remove()...

> As the history of this driver in mainline is convoluted with several
> rounds of fixes it includes the fixes from two commits, namely
> commit 5ff4d5d1af0c ("spi: cadence-quadspi: fix runtime pm disable
> imbalance on probe failure") and commit 5e8bb0cc72f1 ("spi:
> cadence-quadspi: fix runtime pm and clock imbalance on unbind").

Please split this so each patch maps to exactly one upstream commit.

> Fixes: 74b0b4cf13fc ("spi: spi-cadence-quadspi: Fix pm runtime unbalance")
> Cc: stable@vger.kernel.org

Both Fixes: tags in this series point at stable-tree commits rather than
mainline ones.

-- 
Thanks,
Sasha

[PATCH 6.12.y 2/2] spi: cadence-quadspi: release DMA channel on probe failure

[Diogo Ivo (Schneider Electric)]

When spi_register_controller() fails in cqspi_probe(), the DMA channel
acquired via cqspi_request_mmap_dma() is not released, leaking the channel.
Add a dedicated error label to release the DMA channel before the
existing runtime PM teardown path.

This fix is one part of upstream commit f18c8cfa4f1a ("spi: cadence-qspi:
Fix probe error path and remove").

Fixes: b85815675fc5 ("spi: cadence-quadspi: fix cleanup of rx_chan on failure paths")
Cc: stable@vger.kernel.org
Signed-off-by: Diogo Ivo (Schneider Electric) <diogo.ivo@bootlin.com>
---
 drivers/spi/spi-cadence-quadspi.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/spi/spi-cadence-quadspi.c b/drivers/spi/spi-cadence-quadspi.c
index d749dee54015..66304664a1cf 100644
--- a/drivers/spi/spi-cadence-quadspi.c
+++ b/drivers/spi/spi-cadence-quadspi.c
@@ -1982,14 +1982,17 @@ static int cqspi_probe(struct platform_device *pdev)
 	ret = spi_register_controller(host);
 	if (ret) {
 		dev_err(&pdev->dev, "failed to register SPI ctlr %d\n", ret);
-		goto disable_rpm;
+		goto release_dma_chan;
 	}
 
 	pm_runtime_mark_last_busy(dev);
 	pm_runtime_put_autosuspend(dev);
 
 	return 0;
-disable_rpm:
+release_dma_chan:
+	if (cqspi->rx_chan)
+		dma_release_channel(cqspi->rx_chan);
+
 	pm_runtime_put_noidle(dev);
 	pm_runtime_dont_use_autosuspend(dev);
 	pm_runtime_disable(dev);

-- 
2.55.0

Re: [PATCH 6.12.y 0/2] spi: cadence-quadspi: fix probe error path and remove

[Greg Kroah-Hartman]

On Fri, Aug 07, 2026 at 01:49:19PM +0200, Diogo Ivo (Schneider Electric) wrote:
> This series fixes two resource leaks in the cadence-quadspi driver's
> probe error path and remove callback.
> 
> Patch 1 fixes an imbalanced pm_runtime_enable()/pm_runtime_disable()
> pair: when probe defers after cqspi_request_mmap_dma(),
> pm_runtime_disable() is called without a prior pm_runtime_enable().
> It also adds the missing pm_runtime_dont_use_autosuspend() in both
> the probe error path and the remove callback and a missing
> pm_runtime_put_noidle() in the probe error path. This consolidates
> fixes from upstream commits 5ff4d5d1af0c and 5e8bb0cc72f1.
> 
> Patch 2 fixes a DMA channel leak: when spi_register_controller() fails,
> the channel acquired via cqspi_request_mmap_dma() is never released.
> A dedicated error label is added to release it before the runtime PM
> teardown. This fix is one part of upstream commit f18c8cfa4f1a.
> 
> Signed-off-by: Diogo Ivo (Schneider Electric) <diogo.ivo@bootlin.com>
> ---
> Diogo Ivo (Schneider Electric) (2):
>       spi: cadence-quadspi: fix runtime pm cleanup on probe failure/unbind
>       spi: cadence-quadspi: release DMA channel on probe failure
> 
>  drivers/spi/spi-cadence-quadspi.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
> ---
> base-commit: 7351870f56cba79097a2ab57e609469719ea4f73
> change-id: 20260730-cqspi-pm_runtime-a8a1c6122944
> 
> Best regards,
> --  
> Diogo Ivo <diogo.ivo@bootlin.com>
> 


<formletter>

This is not the correct way to submit patches for inclusion in the
stable kernel tree.  Please read:
    https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>